Jeden Skill in Manus ausführen
mit einem Klick

Jeden Skill in Manus mit einem Klick ausführen

noyalib-research-frontier

Sterne0

Forks0

Aktualisiert8. Juli 2026 um 07:42

What's next for noyalib — the open, candidate research directions where this library could advance the state of the art, PLUS the discipline that turns a hunch into an accepted result here. Frontier thesis chosen 2026-07-07: YAML-spec leadership — conformance beyond the official test suite, edge-case authority, upstream spec/suite contributions, becoming the reference implementation for YAML 1.2.2 ambiguities. Load when asking "is there a research angle here?", "what would be worth a paper / RFC / upstream PR?", "how do we know an idea earned its way in?", "what's a falsifiable milestone for a spec-conformance push?". Ambitions are OPEN and CANDIDATE; nothing on this page is a v0.1.0 commitment. Covers the frontier thesis, four open problem statements with first-three-steps grounded in this repo, and the research methodology (evidence bar, hypothesis-predicts-numbers, idea lifecycle) that has produced the results noyalib already ships. Dated 2026-07-07.

Installation

Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.

In Manus ausführen

Quelle

sebastienrousseau

sebastienrousseau/agtmls

GitHub-Repository öffnen Creator-Repositorys ansehen

Download

In Manus ausführen

Verwandte BerufeSOC

Basierend auf der SOC-Berufsklassifikation

SoftwareentwicklerInformatik- und Mathematikberufe·SOC 15-1252

Datei-Explorer

2 Dateien

SKILL.md

readonly

Mehr aus diesem Repository

gleiches Repository

cross-language-port

sebastienrousseau/agtmls

Autonomously port a module, script, CLI, or function from one language to another across the polyglot fleet — Rust, Python, Go, C++, Swift, TypeScript, JavaScript, Ruby, and Bash/Shell — preserving business logic exactly while adopting the target language's idioms, error model, concurrency shape, and toolchain, AND proving behavioural equivalence with a differential golden-I/O harness rather than eyeballing. Load when the user says "port this to <lang>", "translate to <lang>", "rewrite in <lang>", "convert this to <lang>", "reimplement <X> in <lang>", "give me the <lang> equivalent", or asks for a Rust↔Python↔Go↔Swift↔TS/JS↔Ruby↔C++ ↔Bash equivalent of existing code. Not for micro-snippets (do those inline) or behaviour-changing redesigns (that's a rewrite, clarify intent first).

2026-07-210

using-agtmls

sebastienrousseau/agtmls

Meta-router for the AgtMLS skill catalog — read this first when you are unsure which AgtMLS skill (if any) to load for a task, when several skills look like they might apply, or when onboarding to how this hub's skills and shared references fit together. Points at the cross-language-port skill for porting/translating code between languages, at project bundles (e.g. the noyalib/ family) for project-specific work, and at references/ for the shared Definition of Done. Use when the question is "which skill do I use for X" or "how do AgtMLS skills work".

2026-07-210

example-skill

sebastienrousseau/agtmls

Use when the user needs this reusable workflow. Replace this with trigger-rich wording that says when to load the skill, what task it handles, and when not to use it.

2026-07-210

pr-review-and-release

sebastienrousseau/agtmls

Use when preparing a pull request, reviewing implementation risk, writing release notes, checking CI, packaging artifacts, or deciding whether a repository is ready to ship.

2026-07-190

agent-loop-design

sebastienrousseau/agtmls

Use when turning repeated agent work into a loop with trigger conditions, evidence, stopping criteria, uninstall state, validation checks, and reviewable artifacts.

2026-07-190

ai-supply-chain-security

sebastienrousseau/agtmls

Use when auditing AI-agent supply chain risk, prompt or skill provenance, dependency trust, SBOM/SLSA evidence, MCP tool-poisoning, package provenance, or agent runtime guardrails before release.

2026-07-190

name

noyalib-research-frontier

description

noyalib Research Frontier

Progressive disclosure. The bulk of this skill — the four full open-problem statements and the verbatim worked examples — lives in reference.md. This page is the compact index: thesis, problem index, methodology rules, non-goals, cross-refs, provenance.

Open problems + the discipline. Nothing here is committed roadmap; everything is a candidate direction with a falsifiable "you have a result when…" milestone attached.

Audience: mid-level engineer or Sonnet-class model with time to spend on research-shaped work (weeks, not hours).
Scope: pure-Rust YAML library noyalib v0.0.14.
Frontier thesis (chosen 2026-07-07): YAML-spec leadership. Conformance beyond the official test suite; edge-case authority; upstream spec/suite contributions; the reference implementation for YAML 1.2.2 ambiguities.
Date-stamp: 2026-07-07. Every claim below carries a re-verify command so it can be re-checked before being repeated.
Not this skill:
- Positioning / marketing claims → noyalib-external-positioning.
- Executing a measured campaign (coverage, benchmark chase) → noyalib-coverage-campaign.
- Settled battles → noyalib-failure-archaeology.
- Live triage → noyalib-debugging-playbook.

Provenance — re-verify before quoting numbers

Everything volatile below was captured 2026-07-07. Re-run before repeating:

grep -n "skip" crates/noyalib/tests/official_suite.rs | head
grep -n "SKIP_LIST\|skip" crates/noyalib/tests/yaml_compliance_report.rs | head
ls crates/noyalib/tests/yaml-test-suite/*.yaml | wc -l
grep -c '^#\[test\]' crates/noyalib/tests/competitor_bugs.rs
ls fuzz/fuzz_targets/fuzz_diff.rs fuzz/fuzz_targets/fuzz_yaml_v1_1.rs \
    fuzz/fuzz_targets/fuzz_strict.rs

Captured 2026-07-07 (label: candidate, not target):

SKIP_LIST in both official_suite.rs and yaml_compliance_report.rs is &[]. The regression net treats the entire suite as active — any future skip has to be argued for in writing in the same file.
Bundled suite copy: ~351 .yaml case wrappers.
Competitor-bug fixtures: 27 #[test] entries in competitor_bugs.rs.
Differential fuzz targets present: fuzz_diff, fuzz_yaml_v1_1, fuzz_strict, plus fuzz_no_span_loader, fuzz_double_quoted, fuzz_borrowed_alias, fuzz_multi_doc, fuzz_parse, fuzz_roundtrip, fuzz_from_value.

If any has drifted, use the new number — do not quote this page.

Frontier thesis

No Rust YAML crate is treated by the ecosystem as the arbiter of YAML 1.2.2 ambiguities. The seat is empty; noyalib has the minimum viable base camp to compete for it. Five on-the-ground assets justify that claim: a full-suite runner that generates a categorised gap report (not a pass-rate badge); strict-1.2-by-default posture (ADR-0002); a 27-entry competitor_bugs.rs corpus already cross-referenced to peer tickets; a CST (ADR-0001) that can represent byte-exact edge cases; and a fuzz_diff harness already running noyalib × serde_yaml_ng × saphyr. Not "noyalib is the arbiter now" — "the seat is empty and this codebase has the tooling to compete for it".

Full evidence table (all five assets, spec-§ citations, SOTA-failure statement in long form): see reference.md § Frontier thesis — evidence (extended).

Open-problems index

Each is a candidate direction with a first-three-steps block and a falsifiable Result-when milestone. Full narrative for each in reference.md.

(a) Close the skipped-case set — SKIP_LIST = &[] is a published zero; each future skip needs written spec-§ rationale. → reference.md § Open problem (a)
(b) Edge-case conformance database beyond the suite — lift competitor_bugs.rs + failure-archaeology into a citable spec-referenced corpus other crates can adopt. → reference.md § Open problem (b)
(c) Upstream spec/suite contributions — push fail-lenient and fail-value-mismatch cases into yaml-test-suite / spec errata rather than treating the suite as read-only. → reference.md § Open problem (c)
(d) Differential conformance harness as a service — turn fuzz_diff's abort stream into a per-release published divergence table, cited externally. → reference.md § Open problem (d)

Research methodology summary

Compact rule list. Each rule has a specific failure mode in this codebase it's protecting against; the verbatim worked examples that justify each rule live in reference.md § Research methodology — worked examples.

Evidence bar.

One mechanism must explain ALL observations, including negatives. A hypothesis that says "X causes symptom A" but doesn't say why X-adjacent symptom A' is absent is under-specified. → see silent-collapse worked example.
Adversarial refutation before landing: assign someone (Sonnet-class model, or yourself in an adversarial seat) to find the counter-example. Counter-example landing = hypothesis wrong or scope narrower than claimed; both are findings.

Hypothesis predicts numbers.

Benchmark expectations written down before the run. If the number lands where predicted, that's a much stronger signal than "looks fine". → see merge-key clone gate worked example.
If the number lands and the prediction wasn't written first, you don't know whether the mechanism you think explains it actually did.

Idea lifecycle.

Experiment (bench, fuzz target, throwaway branch — cheap, disposable).
Cross-path validation across the ~4-5 parse entry points (from_str, load_all_as, cst::parse_document, Loader, NoSpanLoader).
Change-control PR (spec § or failing test cited; CI green on every feature-matrix job; regression test shipped or written refusal).
Adopted OR documented retirement (fizzled directions → failure- archaeology "Attempted, not adopted").

Where good ideas historically came from. Community PRs (#117, #118) that were narrow + spec-cited + tested; incident follow-ups (every failure-archaeology entry began this way); competitor-bug mining; bench regressions. Full form in reference.md § Where good ideas historically came from.

Non-goals (fenced — do not spend research budget here)

YAML 1.3 / 2.0 speculation. No stable target. 1.2.2 is the active spec; research work targets it.
Forking the yaml-test-suite. Contribute upstream. A private fork loses the "citable common floor" property that makes conforming to the suite meaningful.
Trading the strict-1.2 default for compatibility wins. ADR-0002 is settled; 1.1 semantics remain available via ParserConfig::version.
New feature velocity that isn't security / conformance / performance driven. Per the security-first posture memory (2026-07-07): research budget goes to hardening, correctness, and the frontier thesis, not surface expansion.

When NOT to use this skill

Marketing / positioning claims. "noyalib is the most spec-compliant Rust YAML crate" is a positioning statement, not a research finding. → noyalib-external-positioning (when authored).
Executing a measured campaign (coverage push, benchmark improvement, targeted parity work). → noyalib-coverage-campaign (when authored).
Live triage of a red CI / regression. → noyalib-debugging-playbook.
Looking up why an existing design is shaped a particular way. → noyalib-failure-archaeology (settled battles) or doc/adr/ (settled decisions).
Making a change to code without a research question attached. → noyalib-change-control for the process, noyalib-architecture-contract for the shape constraints.

Re-verify this file

Before quoting anything on this page to an outside audience, re-run the Provenance section commands. The frontier's starting inventory (skip list, suite size, fixture count, differential-target list) is the volatile part — everything else on this page is methodology and thesis. Methodology and thesis don't rot on a weekly cadence, but the numbers do.

Last full re-verify: 2026-07-07.